Isan Traditional Music Dataset

AI Recognition System for Phin and Khaen Playing Styles

🎵 Overview

This project provides a structured, ethically compliant dataset and ML pipeline for training AI models to recognize and analyze the playing styles of two traditional Isan instruments:

• Phin (พิณ): A three-stringed lute from Northeastern Thailand
• Khaen (แคน): A bamboo mouth organ, the national instrument of Laos

🎯 Key Features

• ✅ Ethical Compliance: Full consent, cultural attribution, respectful data handling
• 🎼 Comprehensive Feature Extraction: MFCCs, chroma, spectral, temporal, and pitch features
• 🏗️ Structured Dataset: Organized by instrument and playing technique
• 🤖 ML Pipeline: Complete training and evaluation workflow
• 📊 Detailed Metadata: Cultural context, performer info, technical specifications
• 📚 Documentation: Full reproducibility and transparency

📁 Project Structure

.
├── dataset_manager.py       # Dataset creation and feature extraction
├── model_training.py        # ML training pipeline
├── main.py                  # Demo and visualization
├── dataset/                 # Main dataset directory
│   ├── phin/               # Phin recordings
│   │   ├── fast_plucking/
│   │   ├── legato/
│   │   ├── vibrato/
│   │   └── harmonic_plucking/
│   ├── khaen/              # Khaen recordings
│   │   ├── single_note/
│   │   ├── harmonic_overblow/
│   │   ├── drone/
│   │   └── melodic_phrase/
│   ├── processed_features/  # Extracted features (.npz)
│   ├── annotations/         # Transcriptions
│   └── metadata/           # Documentation
└── models/                 # Trained models

🚀 Quick Start

1. Initialize Dataset Structure

from dataset_manager import IsanInstrumentDataset

# Create dataset infrastructure
dataset = IsanInstrumentDataset("./dataset")
dataset.generate_dataset_description()

2. Add Recordings with Metadata

# Create metadata
metadata = dataset.create_metadata_entry(
    instrument="phin",
    technique="fast_plucking",
    performer="Master Somchai Isan",
    recording_date="2025-11-20",
    location="Sakon Nakhon",
    key="Isan pentatonic scale",
    tempo="♩=80",
    cultural_context="Lam Vang folk song"
)

# Process audio file
features, feature_vector = dataset.process_recording(
    audio_path="path/to/recording.wav",
    metadata=metadata,
    save_features=True
)

3. Train ML Model

python model_training.py

4. Run Demo

python main.py

🎼 Supported Techniques

Phin (พิณ):

• Fast Plucking: Rapid successive plucks
• Legato: Smooth, connected notes
• Vibrato: Pitch modulation
• Harmonic Plucking: Overtone emphasis

Khaen (แคน):

• Single Note: Individual note playing
• Harmonic Overblow: Harmonic overtones
• Drone: Sustained notes
• Melodic Phrase: Melodic sequences

📊 Audio Specifications

• Sample Rate: 44.1 kHz
• Bit Depth: 24-bit
• Format: WAV (uncompressed)
• Contexts: Solo and ensemble performances

🔬 Feature Extraction

Each recording is analyzed for:

• MFCCs (13 coefficients): Timbral characteristics
• Chroma (12 bins): Pitch class distribution
• Spectral Features: Centroid, rolloff, bandwidth
• Temporal: Zero-crossing rate, RMS energy
• Onset Detection: Plucking/breath dynamics
• Pitch Tracking: Melodic contour
• Tempo: Beat tracking

🤝 Ethical Guidelines

1. Consent: All recordings require explicit informed consent
2. Attribution: Performers and communities must be credited
3. Cultural Respect: Document cultural origins and context
4. Privacy: Protect performer identity when requested
5. Usage: Research and education purposes only
6. Compensation: Fair compensation for performers
7. Representation: Include diverse regional styles

📈 Evaluation Metrics

• Accuracy: Overall classification performance
• F1-Score: Especially for rare techniques
• Cross-Cultural: Generalization across regions
• Expert Validation: Feedback from traditional musicians

🛠️ Dependencies

# Install required packages (managed by Replit)
librosa>=0.10.0
numpy>=1.24.0
pandas>=2.0.0
matplotlib>=3.7.0
scikit-learn>=1.3.0

📚 Citation & Usage

When using this dataset:

1. Credit the performers and communities
2. Acknowledge the cultural heritage
3. Reference the Isan/Lao cultural context
4. Use responsibly for research and education
5. Share improvements back to the community

🌏 Cultural Context

This dataset represents musical traditions from:

• Isan Region (Northeastern Thailand)
• Lao PDR (Laos)
• Various regional styles and variations

📖 References

• Thai National Radio Archives
• Isan Music Preservation Society
• Traditional Lao Music Documentation Project

🤔 FAQ

Q: How do I add my own recordings? A: Use dataset.process_recording() with proper metadata and consent documentation.

Q: Can I contribute to the dataset? A: Yes! Ensure ethical guidelines are followed and provide complete metadata.

Q: What if I need to classify a new technique? A: Add the technique to the dataset structure and retrain the model.

📞 Support

For questions or contributions:

• Ensure all ethical guidelines are followed
• Provide complete metadata
• Include consent documentation

---

Dataset created for AI research in traditional music analysis
Respecting cultural heritage and intellectual property rights 🎶